KR20220160660A - Method, apparatus and storage media for removing handwriting from text image - Google Patents

Abstract

The present invention provides a method, apparatus, and storage medium for removing handwritten contents from a text image. A method of removing handwritten content from a text image includes acquiring an input image of a text page to be processed, the input image including a writing area, and the writing area including the writing content; identifying the input image using an image segmentation model to obtain an initial handwritten pixel of the handwritten content; performing a blurring process on the initial handwritten pixel to obtain a handwritten pixel mask area; determining the handwritten content based on a handwritten pixel mask area; and obtaining an output image by removing the handwritten content from the input image.

Description

Method, apparatus and storage media for removing handwriting from text image

The present invention relates to a method, apparatus, and storage medium for removing handwritten contents from a text image.

When the current user captures and scans the text as an image or a file in other formats such as PDF, the current user or other person's handwriting on the original text, such as text marks such as remarks, explanatory characters, comments or notation marks. If the contents already exist, these handwritten contents can also be simultaneously written to the output image or file. If the user does not need the written content or needs to keep the written content confidential, it is relatively difficult for the general user to remove the related written content, and it is not easy to store or distribute. In addition, a text photo taken by a user with a mobile phone may sometimes have shadows on the text photo due to a difference in light irradiation in a shooting environment. If the text photo is printed immediately, the printer prints the shaded portion of the text photo as it is, wasting ink and affecting reading.

In order to solve the above-described defect, the present invention provides a method of removing handwritten content from a text image. Acquiring an input image of a text page to be processed, wherein the input image includes a writing area, and the writing area includes writing contents; identifying the input image using an image segmentation model to obtain an initial handwritten pixel of the handwritten content; performing a blurring process on the initial handwritten pixel to obtain a handwritten pixel mask area; determining the handwritten content in the writing area based on the writing pixel mask area; and obtaining an output image by removing the handwritten content from the input image.

Optionally, in the method for removing handwritten content from a text image provided by the present invention, the step of obtaining an output image by removing the handwritten content from the input image includes the following steps.

A non-writing pixel in the handwritten pixel mask area of the input image is determined based on the pixel value of the initial handwritten pixel and the position of the handwritten pixel mask area. The handwritten pixel mask region content in the input image is removed to obtain an intermediate output image.

Restoring the non-written pixels in the written pixel mask area is performed on the intermediate output image to obtain the output image.

Optionally, in the method for removing handwritten content from a text image provided by the present invention, the step of obtaining an output image by removing the handwritten content from the input image includes the following steps.

A non-writing pixel in the handwritten pixel mask area of the input image is determined based on the pixel value of the initial handwritten pixel and the position of the handwritten pixel mask area.

The output image is obtained by removing the handwritten content in the input image based on the non-written pixels in the handwritten pixel mask area and the handwritten pixel mask area.

Optionally, in the method of removing handwritten content from a text image provided by the present invention, the step of removing the handwritten content from the input image to obtain an output image includes: cutting and removing the writing content from the input image to intermediate obtaining an output image; and performing a binarization process on the intermediate output image to obtain the output image.

Optionally, in the method for removing handwritten content from a text image provided by the present invention, the step of obtaining the output image by removing the handwritten content from the input image includes: acquiring a replacement pixel; and obtaining the output image by removing the handwritten content from the input image by using the replacement pixel to replace the pixel of the handwritten content.

Optionally, in the method of removing handwritten content from a text image provided by the present invention, the output image is obtained by removing the handwritten content from the input image by replacing pixels of the handwritten content using the replacement pixels. The steps may include replacing pixels of the handwritten content using the replacement pixels to remove the handwritten content from the input image to obtain an intermediate output image; and performing a binarization process on the intermediate output image to obtain the output image.

Optionally, in the method of removing handwritten content from a text image provided by the present invention, the replacement pixel is obtained through an image restoration algorithm calculated based on a pixel adjacent area based on a pixel of the handwritten content.

Optionally, in the method for removing handwritten content from a text image provided by the present invention, the acquiring of the replacement pixel further includes acquiring the writing region by identifying the input image using a region identification model. . the replacement pixel is any pixel except for the pixel of the written content in the writing area; Alternatively, the replacement pixel is an average value of pixel values of all pixels except for the pixel of the handwritten content in the writing area.

Optionally, in the method of removing handwritten contents from a text image provided by the present invention, the obtaining of an input image of a text page to be processed comprises: obtaining an original image of the text page to be processed - the original image - contains the text area to be processed; determining the text area to be processed in the original image by performing edge detection on the original image; and obtaining the input image by performing normalization processing on the text area to be processed.

Optionally, in the method for removing handwriting from a text image provided by the present invention, the image segmentation model is a U-Net model for segmenting the pretrained input image.

Optionally, in the method of removing handwritten content from a text image provided by the present invention, a blurring process is performed on the initial handwritten pixel through a Gaussian filter function, a region of the initial handwritten pixel is enlarged, and the handwriting is removed. Get the pixel mask area.

In addition, the present invention further provides an apparatus for removing handwritten contents from a text image. Memory used for non-transitory storage of computer readable instructions; and a processor used to execute the computer readable instructions. When the computer-readable command is executed by the processor, the method of removing handwritten content from a text image according to any one of the above-described embodiments is executed.

In addition, the present invention further provides a storage medium for non-temporarily storing computer readable instructions. When the computer readable command is executed by a computer, the method of removing handwritten content from a text image according to any one of the above-described embodiments may be executed.

BRIEF DESCRIPTION OF THE DRAWINGS To describe the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings in the embodiments. The accompanying drawings in the following description only relate to some embodiments of the invention rather than limit it.
1 is a flowchart of a method of removing handwritten content from a text image according to an embodiment of the present invention.
2A is a schematic diagram of an original image according to an embodiment of the present invention.
2B is a schematic diagram of an output image according to an embodiment of the present invention.
3 is a block diagram of a device for removing written content from a text image according to an embodiment of the present invention.
4 is a schematic diagram of a storage medium according to an embodiment of the present invention.
5 is a schematic diagram of a hardware environment according to an embodiment of the present invention;

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are some, but not all, embodiments of the present invention. Based on the described embodiments of the present invention, all other embodiments obtained by a person skilled in the art without creative efforts fall within the protection scope of the present invention.

Unless defined otherwise, technical or scientific terms used herein shall have ordinary meanings as understood by one of ordinary skill in the art to which this invention belongs. The terms "first", "second" and similar terms used herein do not indicate any order, quantity or importance, but are only used to distinguish different components. "Including" or "inclusive" and similar terms mean that the element or thing preceding the term includes, but does not exclude other elements or things, the elements or things listed after the term and their equivalents. Terms such as “connection” or “interconnection” and similar terms are not limited to physical or mechanical connections and may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are used only when indicating a relative positional relationship. If the absolute position of the description target is changed, the corresponding relative positional relationship may also be changed accordingly.

To keep the following description of embodiments of the present invention clear and concise, the present invention omits detailed descriptions of some well-known functions and known elements.

At least one embodiment of the present invention provides a method, apparatus, and storage medium for removing writing contents from a text image. A method of removing written content from a text image includes acquiring an input image of a text page to be processed, wherein the input image includes a writing area, and the writing area includes the writing content; identifying the input image using an image segmentation model to obtain an initial handwritten pixel of the handwritten content; performing a blurring process on the initial handwritten pixel to obtain a handwritten pixel mask area; determining the handwritten content based on a handwritten pixel mask area; and obtaining an output image by removing handwritten content from the input image.

A method of removing handwritten content from a corresponding text image effectively removes handwritten content within a writing area in an input image, making it easy to output an image or file containing only printed content. In addition, the method of removing handwritten content from a text image may convert an input image into a printable form so that a user can print the input image in paper form and store or distribute the input image.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, but the present invention is not limited to these specific embodiments.

1 is a flowchart of a method of removing written content from a text image according to at least one embodiment of the present invention. 2A is a schematic diagram of an original image according to at least one embodiment of the present invention. 2B is a schematic diagram of an output image according to at least one embodiment of the present invention.

For example, as shown in FIG. 1 , a method for removing handwritten content from a text image provided in an embodiment of the present invention includes steps S10 to S14.

As shown in FIG. 1 , in the method of removing handwritten content from a text image, an input image of a text page to be processed is obtained in step S10.

For example, in step S10, the input image includes a writing area, and the writing area includes writing contents. The input image may be any image including written content.

For example, the input image may be an image captured through an image collection device (eg, a digital camera or mobile phone). The input image may be a grayscale image or a color image. The input image means a form in which a text page to be processed is presented in a visualization manner. For example, a photo of a text page to be processed.

For example, the writing area does not have a fixed shape and is determined according to the writing area. That is, the area where the writing content is located is the writing area. The writing area may have a regular shape (eg, a rectangle) or an irregular shape. The writing area may include a padding area, a draft of writing or other writing display area, and the like.

For example, the input image also includes a text print area, and the text print area includes printed content. The shape of the text printing area may be a regular shape (for example, a rectangle) or an irregular shape. In the embodiment of the present invention, a case where each writing area has a rectangular shape and each text printing area has a rectangular shape will be described as an example. The present invention is not limited to this.

For example, text pages to be processed may include books, newspapers, periodicals, receipts, charts, contracts, and the like. Books, newspapers and periodicals include various file pages with text or graphics. Receipts include various invoices, invoices, courier receipts, and the like. Charts can be of various types, such as year-end statements, personnel records, applications, etc. The contract may include contract text pages in various formats. The present invention does not specifically limit the type of text page to be processed.

For example, the text page to be processed may be text in paper format or may be text in electronic format. For example, when the text page to be processed is a receipt, for example, a delivery receipt, the printed content may include heading text including each item. The handwritten content may include information entered by the user, for example, name, address, phone number, etc. (at this time, the information is personal information entered by the user, not general information). If the text page to be processed is sentence-like text, the printed content may be sentence content, and the handwritten content may be a user's remarks or other handwritten marks. If the text page to be processed is a form such as a personnel record book, the printed content may include item heading characters such as "name", "gender", "ethnicity", and "occupational history". The written contents may include handwritten information such as a user's name, gender (male or female), ethnicity, and occupational history that a user (eg, an employee) writes in a personnel record book. The printed content may further include various symbols, figures, and the like.

For example, the shape of a text page to be processed may be a shape such as a rectangle. The shape of the input image may be a regular shape (eg, parallelogram, rectangle, etc.) to facilitate printing. However, the present invention is not limited thereto. In some embodiments, the input image may be of irregular shape.

For example, since deformation may occur when an image collection device collects an image, the size of an input image and the size of a text page to be processed may be different. However, the present invention is not limited thereto. The size of the input image and the size of the text page to be processed may be the same.

For example, a text page to be processed includes printed content and handwritten content. The printed content is content obtained by printing, and the handwritten content is content written by the user. The writing contents may include writing codes.

"Printed content" indicates content such as characters, symbols, and figures input onto an electronic device through an input device. In addition to this, in some embodiments, when a text page to be processed is text such as a memo, the content of the memo may be handwritten by the user. At this time, the printed content is a printed content on a blank notepad page for writing, for example, a horizontal line.

For example, the printed content may include characters of various languages such as Chinese (eg, Chinese characters or Pinyin), English, Japanese, French, Korean, and the like. In addition, the printed contents may include numbers, various symbols (eg, check marks, insertion marks, various operation codes, etc.) and various figures. The written contents may include letters, numbers, various symbols, and various figures in various languages.

For example, in the example shown in FIG. 2A, the text page 100 to be processed is a form. An area surrounded by four border lines (straight lines 101A to 101D) represents a text area 100 to be processed corresponding to a text page to be processed. In the text area 100 to be processed, the print area includes a form area. The print area may include characters of each item, such as a name, a birthday, and the like. The printed content may include a logo figure (masked) at the top right of the text area 100 to be processed. The writing area includes a writing information area. The handwritten content may include personal information written by the user, for example, a name, birthday information, health information, and a check code written by the user.

For example, the input image may include a plurality of written contents and a plurality of printed contents. A plurality of written contents are spaced apart from each other, and a plurality of printed contents are also spaced apart from each other. For example, some of the written contents among the plurality of written contents may be the same (that is, the written contents have the same sign, but the specific shapes of the written contents are different), and some of the printed contents among the plurality of printed contents may also be the same. have. The present invention is not limited to this. A plurality of written contents may be different from each other, and a plurality of printed contents may also be different from each other.

For example, in some embodiments, step S10 includes acquiring an original image of a text page to be processed, wherein the original image includes a text area to be processed; determining a text area to be processed in the original image by performing edge detection on the original image; and obtaining an input image by performing normalization processing on a text area to be processed.

For example, a text area to be processed can be determined by performing edge detection on the original image by adopting a method such as a neural network or OpenCV-based edge detection algorithm. For example, OpenCV is an open source computer vision library. Edge detection algorithms based on OpenCV include various algorithms such as Sobel, Scarry, Canny, Laplacian, Prewitt, Marr-Hildresh, and scharr.

For example, the step of performing edge detection on the original image to determine the text area to be processed in the original image is the step of processing the original image to obtain a grayscale outline line drawing of the original image - the line drawing is Contains multiple lines - ; obtaining a plurality of initial merged lines by combining similar lines in the diagram, and determining a boundary matrix based on the plurality of initial merged lines; acquiring a plurality of target lines by merging similar lines from the plurality of initial merge lines to obtain target lines, and using the non-merged initial merge lines as target lines; Based on the boundary matrix, determining a plurality of reference boundary lines from the plurality of target lines; acquiring a plurality of boundary regions of a text page to be processed from among the original images by processing the original image through a pretrained boundary region identification model; For each boundary area, determining a target boundary line corresponding to the corresponding boundary line area among a plurality of reference boundary lines; and determining an edge of a text page to be processed among original images based on the determined plurality of target boundary lines.

For example, in some embodiments, the step of processing the original image and obtaining a grayscale outline of the original image includes processing the original image through an OpenCV-based edge detection algorithm, and obtaining a grayscale outline of the original image. It includes obtaining a line.

For example, the step of merging similar lines in the diagram and obtaining a plurality of initial merged lines may include: acquiring a long line in the diagram, wherein the long line has a length exceeding a first predetermined threshold; obtaining a plurality of sets of first type lines from the long lines, wherein the first type lines include at least two sequentially adjacent long lines, and all included angles between any two adjacent long lines are smaller than a second predetermined threshold value. - ; and for each set of first type lines, sequentially merging each long line of the first type lines in the set to obtain one initial merged line.

For example, the boundary matrix is determined according to the following method. That is, a plurality of initial merged lines and lines that are not merged from long lines are re-drafted. The positional information of the pixel points in all the re-drafted lines is mapped to the matrix of the entire original image. A value of a position where a pixel point of such a line is located in a matrix of an original image is set as a first value. A value of a location where a pixel point other than the line is located is set to a second value to form a boundary matrix.

For example, acquiring a target line by merging similar lines from a plurality of initial merged lines may include obtaining a plurality of sets of second type lines from the plurality of initial merged lines - the second type lines are sequentially formed by at least two sets of second type lines. including adjacent initial merging lines, and an included angle between any two adjacent initial merging lines is smaller than a third predetermined threshold value; and for each set of second type lines, sequentially merging each initial merged line in the set of second type lines to obtain one target line.

For example, the first predetermined threshold may be two pixels in length. The second predetermined threshold value and the third predetermined threshold value may be 15 degrees. Note that the first predetermined threshold, the second predetermined threshold, and the third predetermined threshold may be set according to actual application needs.

For example, the step of determining a plurality of reference boundary lines from a plurality of target lines based on the boundary matrix includes the following steps. That is, for each target line, the corresponding target line is extended. After determining one line matrix based on the corresponding extended target line, the corresponding line matrix and boundary matrix are compared. The number of pixel points belonging to the boundary matrix on the extended target line is calculated and used as the score of the target line. That is, by comparing the corresponding line matrix with the boundary matrix, it is determined how many pixel points fall inside the boundary matrix. That is, a score is calculated by determining how many pixel points at the same location among the two matrices have the same first value, for example, 255. Here, the size of the line matrix and the boundary matrix are the same. A plurality of reference boundary lines are determined from the plurality of target lines based on the score of each target line. Note that the number of target lines having excellent scores may be plural. Therefore, based on the score of each target line, a plurality of target lines having the highest score among the plurality of target lines is determined as the reference boundary line.

For example, the line matrix is determined according to the following method. That is, the extended target line or straight line is drawn again. Among the redrawn lines, the location information of the pixel points is mapped to the matrix of the entire original image. In the matrix of the original image, a value of a position where a pixel point of a line is located is set as a first value, and a value of a position where a pixel point other than a line is located is set to a second value to form a line matrix.

For example, the step of determining a target boundary line corresponding to the boundary line region from a plurality of reference boundary lines for each boundary line region includes the following steps. That is, the slope of each reference boundary line is calculated. For each boundary line region, a Hough transform is used to convert the boundary line region into a plurality of straight lines, and an average slope of the plurality of straight lines is calculated. Next, it is determined whether there exists a reference boundary line whose slope and average slope match among the plurality of reference boundary lines. If present, the reference boundary line is determined as the target boundary line corresponding to the boundary line area. When it is determined that there is no reference boundary line whose slope and average slope are matched among the plurality of reference boundary lines, a line matrix formed by the corresponding straight line and a boundary matrix are compared for each straight line obtained by transforming the corresponding boundary line area. The number of pixel points belonging to the boundary matrix on the corresponding straight line is calculated and used as the score of the corresponding straight line. A straight line with the highest score is determined as a target boundary line corresponding to the boundary line area. Here, the size of the line matrix and the boundary matrix are the same. Note that when there are multiple straight lines with the best score, the straight line appearing first among them according to the sorting algorithm is used as the most desirable boundary line.

For example, the boundary area identification model is a neural network-based model. A boundary area identification model can be built by machine learning learning.

For example, after edge detection is performed on the original image, a plurality of target boundary lines (eg, four target boundary lines) may be determined. A text area to be processed is determined by a plurality of target boundary lines. For example, a text area to be processed may be determined based on a plurality of intersections of a plurality of target boundary lines and a plurality of target boundary lines. Each of two adjacent target boundary lines intersects to obtain a difference point, and the plurality of intersection points and the plurality of target boundary lines together define an area of the original image where the text to be processed is located. For example, in the example shown in FIG. 2A , the text area to be processed may be a text area surrounded by four target boundary lines. All four target boundary lines are straight. The four target boundary lines are a first target boundary line 101A, a second target boundary line 101B, a third target boundary line 101C, and a fourth target boundary line 101D. In addition to the text area to be processed, the original image may further include a non-text area. For example, it is an area other than the area surrounded by the four boundary lines in FIG. 2A.

For example, in some embodiments, performing normalization processing on the text area to be processed to obtain an input image includes performing projection transformation on the text area to be processed to obtain a front view of the text area to be processed. Include steps. The corresponding front view is the input image. Perspective Transformation is a technique of projecting a picture onto a new viewing plane, and is also called Projective Mapping. The actual shape of the text to be processed in the original image obtained by shooting is deformed from the original image, resulting in geometric distortion. Like the original image shown in FIG. 2A, the shape of the text to be processed (ie, form) is originally a rectangle. However, the shape of the text to be processed in the original image was deformed and transformed into an irregular polygon. Accordingly, dose transformation is performed on the text area to be processed in the original image. The text area to be processed can be transformed from an irregular polygon to a rectangle or parallelogram, etc. That is, the text area to be processed is normalized to remove the effect of geometric distortion, and a front view of the text to be processed is obtained from the original image. In the projection transformation, a front view of the text to be processed may be obtained by processing pixels in the text area to be processed based on spatial projection conversion coordinates. This will not be repeated here.

In some other embodiments, an image of an independent text area to be processed may be obtained by directly cutting the text area to be processed from an original image without performing normalization on the text area to be processed. The image of the text area to be processed independently is an input image.

For example, the original image may be an image directly collected by an image collection device. It may also be an image obtained after pre-processing an image directly collected by the image collection device. The original image may be a grayscale image or a color image. For example, in order to prevent the effect of data quality, data imbalance, etc. of the original image on removing the handwritten content from the text image, prior to processing the original image, the handwritten content is removed from the text image provided by the embodiment of the present invention. The removal method may further include pre-processing the original image. Pre-processing can make the processing of the original image easier by removing irrelevant or noise information from the original image. Preprocessing may include, for example, processing such as zooming, cropping, gamma correction, image enhancement, or noise reduction filtering for images directly collected by the image collection device.

It should be noted that in some embodiments, an original image may be used as an input image. In this case, for example, the original image is directly identified to determine the written content in the original image, and then the written content in the original image is removed to obtain an output image. Alternatively, the original image is directly identified to determine the written content in the original image, and then the written content in the original image is removed to obtain an intermediate output image. Then, edge detection may be performed on the intermediate output image to determine a text region to be processed in the intermediate output image. Normalization is performed on the text area to be processed to obtain an output image. That is, in some embodiments of the present invention, handwritten content in the original image is first removed to obtain an intermediate output image, and then edge detection and normalization processing may be performed on the intermediate output image.

Subsequently, as shown in FIG. 1 , in step S11, the input image is identified using an image segmentation model to obtain an initial handwritten pixel of the handwritten content.

For example, an image segmentation model represents a model that performs region identification (or segmentation) on an input image. The image segmentation model is implemented by adopting machine learning techniques (eg, convolutional neural network technology) and is executed, for example, on a general-purpose computing device or a dedicated computing device. The image segmentation model is a pretrained model. For example, the neural network applied to the image segmentation model is the same through other neural network models, including deep convolutional neural networks, mask-region convolutional neural networks (Mask-RCNN), deep residual networks, attention models, etc. function can be implemented. This is not limited here.

For example, identifying the input image using an image segmentation model adopts a U-Net model. This is an improved Fully Convolutional Network (FCN) structure, and uses the image semantic segmentation concept of FCN. That is, feature extraction is performed using a convolution layer and a pooling layer, and then the image size is restored using a deconvolution layer. The U-Net network model is a kind of model with relatively excellent image segmentation performance. Deep learning solves the classification problem well. The essence of performing image segmentation using this feature of deep learning is to classify each pixel point in the image. Finally, points of different classes may be displayed using different channels, so that characteristic information in the target area is classified and displayed. An initial handwritten pixel of the handwritten contents may be determined in an input image through a U-Net model. Similarly, the initial handwritten pixel of the handwritten contents may be determined through other neural network models such as Mask-RCNN, for example.

Subsequently, as shown in FIG. 1, in step S12, a blurring process is performed on the initial handwritten pixel to obtain a handwritten pixel mask area. The input image is identified through an image segmentation model. The obtained initial handwritten pixels may not be all handwritten pixels. However, the remaining missing writing pixels are generally all close to the initial writing pixels. Accordingly, a blurring process is performed on the initial handwritten pixel and the handwritten pixel area is enlarged to obtain a handwritten pixel mask area. The handwritten pixel mask area basically includes all handwritten pixels.

For example, a Gaussian blurring process is performed on the initial handwritten pixel through an OpenCV-based Gaussian blur function to enlarge the initial handwritten pixel area. Thus, a handwritten pixel mask area is obtained. The Gaussian filter performs convolution calculations on each point of the input array and the input Gaussian filter template, and then configures these results as a filtered output array. This is a process of performing a weighted average on the image of the initial handwritten pixel, and the value of each pixel point is obtained by weighting the values of itself and other pixels in the adjacent area. After performing the Gaussian blurring process, the handwritten pixel image is blurred but its area is enlarged. For example, a blurring process may be performed on an initial handwritten pixel by adopting another arbitrary blurring technique. This is not limited here.

Subsequently, as shown in FIG. 1 , in step S13, the handwritten content is determined based on the handwritten pixel mask area. The handwritten content is determined by combining initial handwritten pixels based on the handwritten pixel mask area to determine total handwritten pixels of the handwritten content.

Subsequently, as shown in FIG. 1 , in step S14, the handwritten content is removed from the input image to obtain an output image.

For example, in the first relatively preferred embodiment of the present invention, after obtaining the handwritten pixel mask area in step S12, the handwritten pixel mask area may be determined in the input image. Subsequently, non-writing pixels are determined by moving to an area at a corresponding position in the input image. Based on the pixel value of the initial handwritten pixel, another pixel having a relatively large pixel value difference is found in a corresponding region corresponding to the position of the handwritten pixel mask region in the input image and determined as a non-written pixel. For example, a threshold value of a pixel difference value may be set. If there is a pixel whose pixel difference value is outside the threshold range in the area, it is determined as a non-writing pixel.

Then, the content of the handwritten pixel mask area in the input image is removed to obtain an intermediate output image.

For example, the contents of the handwritten pixel mask area can be removed through the OpenCV-based inpaint function. The OpenCV-based inpaint function uses region contiguous to restore a selection in an image. That is, pixels in a corresponding region corresponding to the position of the handwritten pixel mask region in the input image are reconstructed using pixels adjacent to the region. Thus, an effect of removing the handwritten pixel mask region content in the input image is realized, and an intermediate output image is obtained.

Then, restoring non-written pixels in the written pixel mask area is performed on the intermediate output image to obtain the output image.

For example, the pixel value of the non-written pixel in the handwritten pixel mask area of the input image is obtained, and the pixel of the corresponding position point in the intermediate output image is directly replaced. Accordingly, restoration of the non-writing pixel at the corresponding position is completed. Finally, the output image is acquired.

For example, in another preferred embodiment of the present invention, after obtaining the handwritten pixel mask area in step S12, the handwritten pixel mask area may be determined in the input image. Subsequently, non-writing pixels are determined by moving to an area at a corresponding position in the input image. Based on the pixel value of the initial handwritten pixel, another pixel having a relatively large pixel value difference is found in a corresponding region corresponding to the position of the handwritten pixel mask region in the input image and determined as a non-written pixel. For example, a threshold value of a pixel difference value may be set. If there is a pixel whose pixel difference value is outside the threshold range in the area, it is determined as a non-writing pixel.

Subsequently, the output image is obtained by removing the handwritten content in the input image based on the non-written pixels in the handwritten pixel mask area and the handwritten pixel mask area. That is, by excluding non-writing pixels from the writing pixel mask area, pixels in other parts are removed. Therefore, non-writing pixels are preserved so as not to be erroneously removed, and the output image is finally obtained.

For example, the contents of the handwritten pixel mask area of non-written pixels can be excluded through the OpenCV-based inpaint function. The OpenCV-based inpaint function uses region contiguous to restore a selection in an image. That is, pixels other than the non-writing pixels in the corresponding region corresponding to the position of the writing pixel mask region in the input image are reconstructed using pixels adjacent to the region. Therefore, the effect of removing the contents of the handwritten pixel mask area in the input image is realized.

For example, in another relatively preferred embodiment of the present invention, the step of obtaining an output image by removing the handwritten content in the input image is the step of obtaining an intermediate output image by removing the handwritten content from the input image. ; and obtaining an output image by performing a binarization process on the intermediate output image.

The binarization process sets the grayscale values of pixel points on the intermediate output image to 0 or 255. It is also the process of ensuring that the entire intermediate output image exhibits a distinct black-and-white effect. The binarization process can greatly reduce the amount of data in the intermediate output image, thereby emphasizing the outline of the target. The binarization process may convert an intermediate output image into a grayscale image (ie, an output image) having a relatively sharp black-and-white contrast. The noise of the converted grayscale image has relatively little interference, and the discrimination of the content in the output image and the printing effect can be effectively improved.

For example, after cutting and removing the written content from the input image, all pixels in a region corresponding to the written content are removed. That is, there are no blank pixels in the area corresponding to the handwritten content in the input image. When performing the binarization process on the intermediate output image, no processing is performed on areas where pixels are blank in the intermediate output image. Alternatively, when binarization is performed on the intermediate output image, pixels in the intermediate output image may fill a blank area with a gray scale value of 255. Accordingly, the processed text image is formed as a whole, so that an unappealing common area of written contents does not appear.

For example, after binarization is performed on an intermediate output image, an output image is finally obtained so that the user can easily print the corresponding output image in paper form. For example, if the input image is a form, the output image can be printed in paper form and provided to other users to fill in.

For example, the method of binarization processing may be a threshold value method. The threshold method includes the following steps. That is, a binarization threshold is set, and the pixel value of each pixel in the intermediate output image is compared with the binarization threshold. If the pixel value of a specific pixel in the intermediate output image is equal to or greater than the binarization threshold, the pixel value of the corresponding pixel is set to 255 grayscale. If the pixel value of a specific pixel in the intermediate output image is less than the binarization threshold, the pixel value of the corresponding pixel is set to 0 grayscale. Through this, it is possible to perform binarization processing on the intermediate output image.

For example, methods for selecting a binarization threshold include a bimodal method, a P-parameter method, an OTSU method, a maximum entropy method, an iterative method, and the like.

For example, in some embodiments, performing a binarization process on an intermediate output image includes the following steps. That is, an intermediate output image is acquired. Grayscale processing is performed on the intermediate output image to obtain a grayscale image of the intermediate output image. Based on the first threshold, the grayscale image is binarized to obtain a binarized image of the intermediate output image. Using the binarized image as a guide image, guide filtering is performed on the grayscale image to obtain a filter image. High-value pixel points in the filter image are determined according to the second threshold, and the grayscale value of the high-value pixel points is greater than the second threshold. An enlarged image is obtained by performing an enlargement process on a gray scale value of a high-value pixel point based on a predetermined expansion coefficient. Sharpening processing is performed on the extended image to obtain a sharp image. The output image is acquired by adjusting the contrast of the clear image.

For example, methods of gray-scaling processing include component method, maximum value method, average value method, weighted average method, and the like.

For example, the predetermined expansion factor is 1.2 to 1.5, for example 1.3. The grayscale value of each high-value pixel point is multiplied by a predetermined expansion coefficient to perform expansion processing on the grayscale value of the high-value pixel point. Accordingly, an extended image having a clearer black-and-white contrast is obtained.

For example, the second threshold is the sum of the grayscale value of the filter image and the standard difference between the grayscale values.

For example, the step of obtaining a clear image by performing a sharpening process on the enlarged image includes the following steps. That is, a blurring image is acquired by performing a blurring process on the enlarged image by adopting a Gaussian filter. A clear image is obtained by mixing the blurring image and the extended image in proportion based on a predetermined mixing coefficient.

가 확장 이미지의

지점에서의 픽셀 포인트의 그레이스케일 값이라고 가정한다.

는 블러링 이미지의

지점에서의 픽셀 포인트의 그레이스케일 값이다.

는 선명한 이미지의

지점에서의 픽셀 포인트의 그레이스케일 값이다.

은 확장 이미지의 소정 혼합 계수이다.

는 블러링 이미지의 소정 확장 계수이다.

,

,

는 하기 관계를 충족한다.for example,

of the extended image

It is assumed to be the grayscale value of the pixel point at the point.

of the blurring image

The grayscale value of the pixel point at the point.

of the clear image

The grayscale value of the pixel point at the point.

is a predetermined blending coefficient of the extended image.

is a predetermined scaling factor of the blurring image.

,

,

satisfies the following relationship.

For example, the predetermined blending coefficient of the enlarged image is 1.5, and the predetermined blending factor of the blurred image is -0.5.

For example, adjusting the contrast of the sharp image includes adjusting a gray scale value of each pixel point of the sharp image based on an average gray scale value of the sharp image.

For example, the gray scale value of each pixel point of a sharp image can be adjusted through the following formula.

은 개선된 이미지의

지점에서의 픽셀 포인트의 그레이스케일 값이다.

은 선명한 이미지의 그레이스케일 평균값이다.

는 선명한 이미지의

지점에서의 픽셀 포인트의 그레이스케일 값이고,

는 강도 값이다. 예를 들어, 강도 값은 0.1 내지 0.5일 수 있다. 예를 들어 강도 값은 0.2일 수 있다. 실제 응용에서 강도 값은 최종 도달해야 하는 흑백 개선 효과에 따라 선택할 수 있다.From here

of the improved image

The grayscale value of the pixel point at the point.

is the grayscale average value of the clear image.

of the clear image

is the grayscale value of the pixel point at the point,

is the strength value. For example, the intensity value may be between 0.1 and 0.5. For example, the intensity value may be 0.2. In actual application, the intensity value can be selected according to the black-and-white enhancement effect to be finally reached.

For example, as shown in Fig. 1, step S14 includes obtaining a replacement pixel; and removing the handwritten content from the input image by replacing pixels of the written content using the replacement pixels to obtain an output image.

For example, the replacement pixel may be an adjacent pixel outside the handwritten pixel mask area. That is, it is an adjacent pixel outside the handwritten pixel mask area of the handwritten pixel that is currently to be replaced. Similarly, pixel replacement processing can be performed directly using the OpenCV-based inpaint function.

For example, a handwritten pixel replacement process may be performed by adopting an area identification method. First, a writing region is acquired through a region identification model. The replacement pixel may be a pixel value of any one pixel excluding the pixel of the handwritten content in the writing area. Alternatively, the replacement pixel is an average value (for example, a geometric average value) of pixel values of all pixels in the writing area excluding pixels of the written content. Alternatively, the replacement pixel value may be a fixed value, such as a 255 grayscale value for example. For example, by using an image segmentation model such as a U-Net model, a replacement pixel may be obtained by directly extracting any one pixel except for a handwriting content pixel in the writing area. Alternatively, it should be noted that replacement pixels may be obtained based on the pixel values of all pixels after extracting all pixels except for the pixels of the handwritten content in the writing area using an image segmentation model such as the U-Net model.

For example, the step of obtaining an output image by removing the handwritten content from an input image by replacing pixels of the handwritten content using the replacement pixels may include replacing pixels of the handwritten content using the replacement pixels to remove the handwritten content from the input image. removing to obtain an intermediate output image; and obtaining an output image by performing a binarization process on the intermediate output image.

For a description of performing region identification, binarization processing, etc. on the region identification model, reference may be made to the above related description. The overlapping parts will not be repeated here.

For example, an output image as shown in FIG. 2B may be acquired after processing to remove handwritten content from a text image is performed on the original image shown in FIG. 2A. The output image is a binarized image. As shown in FIG. 2B , all handwritten contents are removed from the corresponding output image to obtain an empty form without user input information.

In an embodiment of the present invention, a model (eg, an arbitrary model such as a region identification model or an image segmentation model) may not be a simple mathematical model, but may be a module for receiving input data, executing data processing, and outputting a processing result. The module may be implemented as a software module, a hardware module (eg, a hardware neural network), or a combination of software and hardware. In some embodiments, the region identification model and/or image segmentation model includes code and programs stored in memory. The processor may implement some or all functions of the region identification model and/or image segmentation model as described above by executing the above codes and programs. In some other embodiments, a region identification model and/or an image segmentation model may be used to implement the functions described above by including one circuit board or a combination of multiple circuit boards. In some embodiments, the circuit board or combination of circuit boards may include (1) one or more processors; (2) one or more non-transitory computer readable memories associated with the processor; and (3) firmware stored in memory executable by the processor.

Note that, in an embodiment of the present invention, before obtaining an input image, the method for removing handwritten content from a text image further includes a learning step. The learning step includes a process of training a region identification model and an image segmentation model. Note that the region identification model and the image segmentation model may be trained separately, or the region identification model and the image segmentation model may be trained simultaneously.

For example, a region to be learned through a first sample image in which a text print area (eg, the number of displayed text print regions is at least one) and a writing area (eg, the number of displayed text print regions is at least one) are displayed. A region identification model may be acquired by performing learning on the identification model. For example, the learning process of the region identification model to be learned includes acquiring the region identification model by learning the region identification model to be learned using a plurality of first sample images in which text printing regions and writing regions are displayed in the learning step. can do.

For example, the step of learning a region identification model to be learned using several first sample images includes the following steps. That is, a current first sample image is acquired from several first sample images. The current first sample image is processed using the region to be learned identification model to obtain a learning text print region and a learning writing region. A first loss value of a region identification model to be learned is calculated through a first loss function based on the displayed text printing area and writing area, the learning text printing area, and the learning writing area of the current first sample image. Based on the first loss value, a parameter of a region identification model to be learned is modified. When the first loss function satisfies the first predetermined condition, a region identification model that has been learned is acquired. If the first loss function does not satisfy the first predetermined condition, the above-described learning process is executed again by continuously inputting the first sample image.

For example, in one example, the above-described first predetermined condition is loss convergence of the first loss function when a predetermined number of first sample images are input (ie, the first loss value does not significantly decrease any more). respond to For example, in another example, the above-described first predetermined condition is that the number of times of learning or the learning period reaches a predetermined number (eg, the predetermined number may be millions).

For example, an image segmentation model may be obtained by learning an image segmentation model to be learned through a second sample image displaying handwriting content pixels. When displaying the handwritten content pixels in the second sample image, the second sample image may be enlarged to accurately display all the handwritten content pixels. An image segmentation model is built by performing machine learning based on various handwriting features (eg, pixel grayscale features, font features, etc.).

For example, the learning process of the image segmentation model to be trained may include acquiring the image segmentation model by learning the image segmentation model to be trained using a plurality of second sample images on which handwriting content pixels are displayed in the learning step. .

For example, the step of learning a region identification model to be learned using several second sample images includes the following steps. That is, a current second sample image is obtained from several second sample images. The current second sample image is processed using the image segmentation model to be learned to obtain training handwritten content pixels. A second loss value of an image segmentation model to be trained is calculated through a second loss function based on the handwriting content pixels displayed in the current second sample image and the training handwriting content pixels. Parameters of an image segmentation model to be learned are modified based on the second loss value. When the second loss function satisfies the second predetermined condition, a trained image segmentation model is obtained. If the second loss function does not satisfy the second predetermined condition, the above-described learning process is repeatedly executed by continuously inputting the second sample image.

For example, in one example, the above-described second predetermined condition is that the loss convergence of the second loss function when a certain number of second sample images are input (ie, the second loss value no longer significantly decreases). respond to For example, in another example, the above-described second predetermined condition is that the number of times of learning or the learning period reaches a predetermined number (eg, the predetermined number may be millions).

For those skilled in the art, several first training sample images and several second training sample images may be the same or different.

At least one embodiment of the present invention further provides an apparatus for removing handwritten content from a text image. 3 is a block diagram of a device for removing writing content from a text image provided by at least one embodiment of the present invention.

As shown in FIG. 3 , an apparatus 300 for removing handwritten contents from the text image includes a processor 302 and a memory 301 . The assembly of the device 300 for removing handwritten content from a text image shown in FIG. 3 is illustrative and non-limiting. According to actual application needs, the device 300 for removing handwritten content from the text image may further include other assemblies. For example, memory 301 is used to non-transitory store computer readable instructions. Processor 302 is used to execute computer readable instructions. When the computer readable instruction is executed by the processor 302, the method of removing handwritten content from a text image according to any one of the above-described embodiments is executed.

The apparatus 300 for removing handwritten contents from a text image provided by an embodiment of the present invention may be used to implement a method of removing written contents from a text image provided by an embodiment of the present invention. The device 300 for removing handwritten content from the text image may be configured on an electronic device. The electronic device may be a personal computer, a mobile terminal, and the like. The mobile terminal may be a hardware device having various operating systems such as a mobile phone and a tablet PC.

For example, as shown in FIG. 3 , the device 300 for removing handwritten content from a text image may further include an image acquisition member 303 . The image acquisition member 303 is used to acquire a text image. For example, an image of paper text is acquired. Memory 301 may further be used to store text images. The processor 302 is further used to obtain an input image by reading and processing the text image. The text image may be the original image described in the above-described method of removing handwritten content from the text image.

For example, the image acquisition member 303 is the image collection device described in the above-described method for removing handwritten content from a text image. For example, the image acquisition member 303 may be a camera of a smart phone, a camera of a tablet PC, a camera of a personal computer, a lens of a digital camera, a network camera, and other devices usable for collecting images.

For example, in the embodiment shown in FIG. 3 , the image acquiring member 303 , the memory 301 , the processor 302 , and the like may be physically integrated into the same electronic device. The image acquisition member 303 may be a camera configured on an electronic device. The memory 301 and the processor 302 receive an image transmitted from the image acquisition member 303 via an internal bus. Also, for example, the image acquisition member 303 and the memory 301/processor 302 may be configured separately on a physical location. The memory 301 and the processor 302 may be integrated in the first user's electronic device (eg, the first user's computer, mobile phone, etc.). The image acquisition member 303 may be integrated into an electronic device of a second user (the first user and the second user are different). The electronic device of the first user and the electronic device of the second user may be configured separately on a physical location. Also, communication may be performed between the electronic device of the first user and the electronic device of the second user by wired or wireless methods. In other words, after the image acquisition member 303 on the electronic device of the second user collects the original image, the electronic device of the second user may transmit the original image to the electronic device of the first user in a wired or wireless manner. . The electronic device of the first user receives the original image and performs subsequent processing on the original image. For example, memory 301 and processor 302 may be integrated in a cloud server. The cloud server receives the original image and processes the original image.

For example, the device 300 for removing handwritten content from a text image may further include an output device. An output device is used to output the output image. For example, the output device may include a display (eg, an organic light emitting diode display, a liquid crystal display), an imaging device, and the like. Displays and imaging devices may be used to display output images. An output device may include a printer. A printer is used to print the output image.

For example, communication may be performed between assemblies such as the processor 302 and the memory 301 through a network connection. The network may include a wireless network, a wired network, and/or any combination of wireless and wired networks. The network may include a local area network, the Internet, a communication network, the Internet of Things based on the Internet and/or a communication network, and/or any combination of the above networks. A wired network may perform communication by employing, for example, twisted pair line, coaxial cable, or optical fiber transmission. The wireless network may adopt a communication method such as a 3G/4G/5G mobile communication network, Bluetooth, Zigbee, or WiFi, for example. The present invention is not limited to the type and function of the network here.

For example, the processor 302 may control other assemblies of the device 300 that remove handwriting from a text image to execute a desired function. The processor 302 may be a device having data processing capability and/or program execution capability, such as a central processing unit (CPU), a tensor processor (TPU), or a graphics processor (GPU). The central processing unit (CPU) may be of X86 or ARM architecture or the like. The GPU can be independently integrated directly on the motherboard or embedded in a chip in the north bridge of the motherboard. A GPU may be embedded on a central processing unit (CPU).

For example, memory 301 may include any combination of one or more computer program products. A computer program product may include various forms of computer readable storage media, such as volatile memory and/or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and/or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable and programmable read-only memory (EPROM), portable compact disk read-only memory (CD-ROM), USB memory, flash memory, etc. have. One or more computer readable instructions may be stored on the computer readable storage medium. The processor 302 may implement various functions of the device 300 for removing handwritten contents from a text image by executing the computer readable instructions. The storage medium may further store various application programs and various data.

The apparatus 300 for removing handwritten content from a text image may refer to the relevant description of the embodiment of the method for removing handwritten content from a text image for a detailed description of a process of executing a method of removing written content from a text image. The overlapping parts will not be repeated here.

At least one embodiment of the invention further provides a storage medium. 4 is a schematic diagram of a storage medium provided by at least one embodiment of the present invention. For example, as shown in FIG. 4 , one or more computer readable instructions 501 may be non-temporarily stored on a storage medium 500 . For example, when the computer readable instructions 501 are executed by a computer, one or more steps of the above-described method of removing handwritten contents from a text image may be executed.

For example, the storage medium 500 may be applied to the device 300 for removing handwritten contents from the text image. For example, this may include a memory 301 in the device 300 that removes handwriting from a text image.

For example, a description of the storage medium 500 may refer to a description of a memory among embodiments of the device 300 for removing handwritten content from a text image. The overlapping parts will not be repeated here.

5 is a schematic diagram of a hardware environment according to at least one embodiment of the present invention. A device for removing handwritten content from a text image provided by an embodiment of the present invention may be applied to an Internet system.

A device for removing writing contents from a text image mentioned in the present invention can be implemented using the computer system provided in FIG. 5 . Computer systems of this type include personal computers, notebook computers, tablet PCs, mobile phones and any smart device. In this embodiment, a specific system analyzes a hardware platform including a user interface using a functional block diagram. Such a computer system may include a general purpose computer device or a special purpose computer device. Both computer devices can be used to implement the device for removing handwritten content from text images in this embodiment. A computer system may implement any assembly of the information necessary to implement the presently described method of removing handwritten content from an image of text. For example, a computer system may be implemented by a computer device by its hardware device, software program, firmware, and other combinations. For ease of use, Figure 5 only depicts a computer device. However, the related computer functions of the information required to implement the method for removing handwritten content from text images described in this embodiment are implemented by a set of similar platforms in a distributed manner. This distributes the processing load of the computer system.

As shown in FIG. 5 , the computer system may include a communications port 250 . Connected to this is a network that implements data communication. For example, the communication port 250 may communicate with the image acquisition member 403 described above. The computer system may further include a processor set 220 (ie, the processor described above) for executing program instructions. The processor set 220 may include at least one processor (eg, CPU). The computer system may include one internal communication bus 210 . A computer system may include program storage units and data storage units (ie, the memory or storage media described above) in different formats. For example, hard disk 270 , read-only memory (ROM) 230 , random access memory (RAM) 240 may be used in various data files and processor set 220 used for computer processing and/or communication. It can be used to store program instructions executable by The computer system may include one input/output assembly 260 . Input/output assembly 260 may support input/output data flow between a computer system and another assembly (eg, user interface 280, which may be a display as described above). . The computer system may send and receive information and data through communication port 250 .

In some embodiments, the computer system described above may be used to configure a server in an Internet communication system. A server of an Internet communication system may be a server hardware device or a group of servers. Each server in the server group can be connected by a wired or wireless network. Server groups can be centralized, such as in a data center. Server groups can also be distributed, such as distributed systems.

Each block in the block diagram and/or flow diagram, and the combination of blocks in the block diagram and/or flow diagram of the present invention, can be implemented as a dedicated hardware-based system that executes specified functions or operations. Alternatively, it should be noted that it may be implemented as a combination of dedicated hardware and computer program instructions. It is known to those skilled in the art that hardware implementation, software implementation, and software/hardware combination are all equivalent.

The present invention needs to further explain the following several things.

(1) The accompanying drawings of the embodiments of the present invention are merely structures related to the embodiments of the present invention, and other structures may refer to conventional designs.

(2) For clarity, the thickness and size of layers or structures are enlarged in the accompanying drawings for explaining the embodiments of the present invention. When an element such as a layer, film, region or substrate is positioned “above” or “below” another element, that element may be positioned “directly” “above” or “below” the other element, or intervening elements may exist. can understand that there is

(3) As long as there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

The above is merely a specific implementation mode of the present invention, and the scope of protection of the present invention is not limited thereto. The protection scope of the present invention should be based on the protection scope of the claims.

Claims (13)

A method for removing handwriting from a text image,
acquiring an input image of a text page to be processed, wherein the input image includes a writing area, and the writing area includes writing content;
identifying the input image using an image segmentation model to obtain an initial handwritten pixel of the handwritten content;
performing a blurring process on the initial handwritten pixel to obtain a handwritten pixel mask area;
determining the handwritten content in the writing area based on the writing pixel mask area; and
and removing the handwritten content from the input image to obtain an output image. According to claim 1,
Obtaining an output image by removing the handwritten content from the input image,
determine non-written pixels in the handwritten pixel mask area in the input image based on the pixel value of the initial handwritten pixel and the position of the handwritten pixel mask area, and remove the content of the handwritten pixel mask area in the input image; obtaining an intermediate output image; and
and performing restoration of non-written pixels in the handwritten pixel mask area on the intermediate output image to obtain the output image. According to claim 1,
Obtaining an output image by removing the handwritten content from the input image,
determining a non-writing pixel in the handwritten pixel mask area in the input image based on the pixel value of the initial handwritten pixel and the position of the handwritten pixel mask area; and
and obtaining the output image by removing the handwritten content in the input image based on the non-written pixels in the handwritten pixel mask area and the handwritten pixel mask area. How to. According to claim 1,
Obtaining an output image by removing the handwritten content from the input image,
obtaining an intermediate output image by cutting and removing the handwritten content from the input image; and
and obtaining the output image by performing binarization on the intermediate output image. According to claim 1,
Obtaining the output image by removing the handwritten content from the input image,
obtaining replacement pixels; and
and removing the handwritten content from the input image by replacing pixels of the handwritten content using the replacement pixels to obtain the output image. According to claim 5,
Obtaining the output image by removing the handwritten content from the input image by using the replacement pixel to replace the pixel of the handwritten content,
obtaining an intermediate output image by removing the handwritten content from the input image by replacing pixels of the handwritten content using the replacement pixels; and
and obtaining the output image by performing binarization on the intermediate output image. According to claim 5,
The method of claim 1 , wherein the replacement pixels are obtained through an image restoration algorithm calculated based on pixel adjacent areas according to pixels of the handwritten content. According to claim 5,
The step of acquiring the replacement pixel further includes acquiring the writing area by identifying the input image using a region identification model, wherein the replacement pixel is any pixel in the writing area excluding the pixel of the handwritten content. is; or
Wherein the replacement pixel is an average value of pixel values of all pixels in the writing area except for the pixel of the writing content. According to any one of claims 1 to 8,
The step of acquiring an input image of the text page to be processed,
obtaining an original image of the text page to be processed, wherein the original image includes a text area to be processed;
determining the text area to be processed in the original image by performing edge detection on the original image; and
and obtaining the input image by performing normalization processing on the text area to be processed. According to claim 1,
Wherein the image segmentation model is a pretrained U-Net model for segmenting the input image. According to claim 1,
The method of claim 1 , further comprising performing a blurring process on the initial handwritten pixel through a Gaussian filter function and enlarging an area of the initial handwritten pixel to obtain the handwritten pixel mask area. An apparatus for removing handwriting from a text image, comprising:
memory used for non-transitory storage of computer readable instructions; and
A method comprising a processor used to execute the computer readable instructions, wherein when the computer readable instructions are executed by the processor, removing handwritten contents from a text image according to any one of claims 1 to 11. A device for removing handwriting from a text image, characterized in that the execution. A storage medium for non-temporarily storing computer readable instructions,
A storage medium capable of executing the method of removing writing contents from a text image according to any one of claims 1 to 11 when the computer readable command is executed by a computer.

Images